Fix endless loop in get_user_groups

[GuillaumeGomez]

Fixes #1145.

Can you confirm it fixes the issue for you @aw-cf please?

[aw-cf]

I would suggest to revert those changes again and fix this by just adding groups.set_len(nb_groups as _); before calling reserve.

Here is what the full function would look like:

pub(crate) unsafe fn get_user_groups(
    name: *const libc::c_char,
    group_id: libc::gid_t,
) -> Vec<Group> {
    let mut buffer = Vec::with_capacity(2048);
    let mut groups = Vec::with_capacity(256);

    loop {
        let mut nb_groups = groups.capacity();
        if getgrouplist(
            name,
            group_id as _,
            groups.as_mut_ptr(),
            &mut nb_groups as *mut _ as *mut _,
        ) == -1
        {
            // Ensure the length matches the number of returned groups.
            // Needs to be updated for `Vec::reserve` to actually add additional capacity.
            groups.set_len(nb_groups as _);
            groups.reserve(256);
            continue;
        }
        groups.set_len(nb_groups as _);
        return groups
            .iter()
            .filter_map(|group_id| {
                let name = get_group_name(*group_id as _, &mut buffer)?;
                Some(Group {
                    name,
                    id: Gid(*group_id as _),
                })
            })
            .collect();
    }
}

[aw-cf]

reserve is not doing anything here because it will only act when additional > .capacity() - .len().
So the key change that is missing here is to modify the length via groups.set_len(nb_groups as _) before calling reserve() .

I recommend adding some logging before calling reserve and setting REALLOC_SIZE = 1 to make the issue more clear.

            println!("before reserve: capacity: {} length: {}", groups.capacity(), groups.len());
            groups.reserve(1);
            println!("after reserve: capacity: {} length: {}", groups.capacity(), groups.len());

The current change is causing memory access violation because nb_groups gets larger than the capacity of groups.

[GuillaumeGomez]

Oh I see. I completely misunderstood the issue... Thanks for the explanations. Don't you prefer to send a PR? At this point you already did all the work so you'd get ownership of the commit as well.

[aw-cf]

No worries and please feel free to push the changes yourself given you have the PR up already.

[GuillaumeGomez]

I copied your code and I added you as co-author (in an incomplete way as I don't have your email). If you want me to add the email, don't hesitate to send it to me, otherwise if you don't want, I'll just merge as is.

Thanks a lot again!

[aw-cf]

Thank you for getting this fixed right away!

[GuillaumeGomez]

It'll be part of the next major release though. I'm hoping to have it finished in the next weeks.